app.py

import spaces
import transformers
import re
from transformers import AutoTokenizer, pipeline
import torch
import html
import gradio as gr
import tempfile
import os
import pandas as pd

# Define the device
device = "cuda" if torch.cuda.is_available() else "cpu"

# Load models
editorial_model = "LLMDH/Estienne"
bibliography_model = "PleIAs/Bibliography-Formatter"
bibliography_style = "PleIAs/Bibliography-Classifier"

tokenizer = AutoTokenizer.from_pretrained(editorial_model, model_max_length=512)

editorial_classifier = pipeline(
    "token-classification", model=editorial_model, aggregation_strategy="simple", device=device
)
bibliography_classifier = pipeline(
    "token-classification", model=bibliography_model, aggregation_strategy="simple", device=device
)

style_classifier = pipeline("text-classification", model=bibliography_style, tokenizer=tokenizer, device=device)

# Helper functions
def preprocess_text(text):
    text = re.sub(r'<[^>]+>', '', text)
    text = re.sub(r'\n', ' ', text)
    text = re.sub(r'\s+', ' ', text)
    return text.strip()

def split_text(text, max_tokens=500):
    parts = text.split("\n")
    chunks = []
    current_chunk = ""

    for part in parts:
        temp_chunk = current_chunk + "\n" + part if current_chunk else part
        num_tokens = len(tokenizer.tokenize(temp_chunk))

        if num_tokens <= max_tokens:
            current_chunk = temp_chunk
        else:
            if current_chunk:
                chunks.append(current_chunk)
            current_chunk = part

    if current_chunk:
        chunks.append(current_chunk)

    if len(chunks) == 1 and len(tokenizer.tokenize(chunks[0])) > max_tokens:
        long_text = chunks[0]
        chunks = []
        while len(tokenizer.tokenize(long_text)) > max_tokens:
            split_point = len(long_text) // 2
            while split_point < len(long_text) and not re.match(r'\s', long_text[split_point]):
                split_point += 1
            if split_point >= len(long_text):
                split_point = len(long_text) - 1
            chunks.append(long_text[:split_point].strip())
            long_text = long_text[split_point:].strip()
        if long_text:
            chunks.append(long_text)

    return chunks

def disambiguate_bibtex_ids(bibtex_entries):
    id_count = {}
    disambiguated_entries = []

    for entry in bibtex_entries:
        # Extract the current ID
        match = re.search(r'@\w+{(\w+),', entry)
        if not match:
            disambiguated_entries.append(entry)
            continue

        original_id = match.group(1)
        
        # Check if this ID has been seen before
        if original_id in id_count:
            id_count[original_id] += 1
            new_id = f"{original_id}{chr(96 + id_count[original_id])}"  # 'a', 'b', 'c', etc.
            new_entry = re.sub(r'(@\w+{)(\w+)(,)', f'\\1{new_id}\\3', entry, 1)
            disambiguated_entries.append(new_entry)
        else:
            id_count[original_id] = 0
            disambiguated_entries.append(entry)

    return disambiguated_entries

def remove_punctuation(text):
    return re.sub(r'[^\w\s]', '', text)

def extract_year(text):
    year_match = re.search(r'\b(\d{4})\b', text)
    return year_match.group(1) if year_match else None

def create_bibtex_entry(data):
    if 'journal' in data:
        entry_type = 'article'
    elif 'booktitle' in data:
        entry_type = 'inproceedings'
    else:
        entry_type = 'book'

    none_content = data.pop('none', '')
    year = extract_year(none_content)
    if year and 'year' not in data:
        data['year'] = year

    if "year" in data:
        match_year = re.search(r'(\d{4})', data['year'])
        
        if match_year:
            data['year'] = match_year.group(1)
            year = data['year']
        else:
            data.pop('year', '')

    #Pages conformity.
    if 'pages' in data:
        match = re.search(r'(\d+(-\d+)?)', data['pages'])
        if match:
            data['pages'] = match.group(1)
        else:
            data.pop('pages', '')

    author_words = data.get('author', '').split()
    first_author = author_words[0] if author_words else 'unknown'
    bibtex_id = f"{first_author}{year}" if year else first_author
    bibtex_id = remove_punctuation(bibtex_id.lower())

    bibtex = f"@{entry_type}{{{bibtex_id},\n"
    for key, value in data.items():
        if value.strip():
            if key in ['volume', 'year']:
                value = remove_punctuation(value)
            if key == 'pages':
                value = value.replace('p. ', '')
            if key != "separator":
                bibtex += f"  {key.lower()} = {{{value.strip()}}},\n"
    bibtex = bibtex.rstrip(',\n') + "\n}"
    return bibtex

def save_bibtex(bibtex_content):
    with tempfile.NamedTemporaryFile(mode='w', delete=False, suffix='.bib') as temp_file:
        temp_file.write(bibtex_content)
    return temp_file.name

class CombinedProcessor:
    @spaces.GPU(duration=120)
    def process(self, user_message):
        #Precaution to reinforce bibliography detection.
        editorial_text = "Bibliography\n" + user_message

        #Our fix for the lack of newline in deberta
        editorial_text = re.sub("\n", " ¶ ", editorial_text)
        
        print(editorial_text)
        num_tokens = len(tokenizer.tokenize(editorial_text))
        
        batch_prompts = split_text(editorial_text, max_tokens=500) if num_tokens > 500 else [editorial_text]
    
        editorial_out = editorial_classifier(batch_prompts)
        editorial_df = pd.concat([pd.DataFrame(classification) for classification in editorial_out])
        
        # Filter out only bibliography entries
        bibliography_entries = editorial_df[editorial_df['entity_group'] == 'bibliography']['word'].tolist()
        
        bibtex_entries = []
        list_style = []
        
        for entry in bibliography_entries:
            print(entry)
            entry = re.sub(r'- ?[\n¶] ?', r'', entry)
            entry = re.sub(r' ?[\n¶] ?', r' ', entry)
            style = pd.DataFrame(style_classifier(entry, truncation=True, padding=True, top_k=1))
            list_style.append(style)
            entry = re.sub(r'\s*([;:,\.])\s*', r' \1 ', entry)
            #print(entry)
            bib_out = bibliography_classifier(entry)
            bib_df = pd.DataFrame(bib_out)
            
            bibtex_data = {}
            current_entity = None
            for _, row in bib_df.iterrows():
                entity_group = row['entity_group']
                word = row['word']
                
                if entity_group != 'None':
                    if entity_group in bibtex_data:
                        bibtex_data[entity_group] += ' ' + word
                    else:
                        bibtex_data[entity_group] = word
                    current_entity = entity_group
                else:
                    if current_entity:
                        bibtex_data[current_entity] += ' ' + word
                    else:
                        bibtex_data['None'] = bibtex_data.get('None', '') + ' ' + word
            
            bibtex_entry = create_bibtex_entry(bibtex_data)
            bibtex_entries.append(bibtex_entry)

        list_style = pd.concat(list_style)
        list_style = list_style.groupby('label')['score'].mean().sort_values(ascending=False).reset_index()
        top_style = list_style.iloc[0]['label']
        top_style_score = list_style.iloc[0]['score']

        # Create the style information string
        style_info = f"Estimated bibliographic style: {top_style} (Mean score: {top_style_score:.6f})"

        # Join BibTeX entries
        bibtex_content = "\n\n".join(bibtex_entries)

        return style_info, bibtex_content

# Create the processor instance
processor = CombinedProcessor()

# Define the Gradio interface
with gr.Blocks(theme='JohnSmith9982/small_and_pretty') as demo:
    gr.HTML("""<h1 style="text-align:center">Reversed Zotero</h1>""")
    text_input = gr.Textbox(label="Your unstructured bibliography", type="text", lines=10)
    text_button = gr.Button("Process Text")
    style_output = gr.Textbox(label="Bibliographic Style", lines=2)
    bibtex_output = gr.Textbox(label="BibTeX Entries", lines=15)
    
    export_button = gr.Button("Export BibTeX")
    export_output = gr.File(label="Exported BibTeX File")

    text_button.click(processor.process, inputs=text_input, outputs=[style_output, bibtex_output])
    export_button.click(save_bibtex, inputs=[bibtex_output], outputs=[export_output])

if __name__ == "__main__":
    demo.queue().launch()